mirror of
https://github.com/stablyai/orca.git
synced 2026-09-22 08:02:28 +00:00
Dense SGR pacing made flushTerminalOutput() stop early — it refused to write while a batch was in flight and stopped after one 4 KiB batch. The replay and shutdown-capture callers pass no char budget precisely because they write straight to xterm next, so a partial flush let stale queued bytes land on top of a freshly restored screen. Treat a flush with no maxChars as an ordering barrier: skip both pacing exits and drain the whole queue. Budgeted callers keep dense pacing.